Cross-component Transfer Fault Diagnosis via Joint Distribution Adaptation on Envelope Spectrum Features

In recent years, multiple domain adaptation methods have been proposed to improve the performance of fault diagnosis models on cross-component fault diagnosis tasks. However, most of these methods only focus on marginal distribution alignment or conditional distribution alignment between cross-domain fault features, ignoring the weight relationship between the two, resulting in the problem of poor performance of fault diagnosis models in target scenarios. In order to improve the fault diagnosis performance of the fault diagnosis model on cross-component tasks, this paper proposes a cross-component transfer fault diagnosis method via envelope spectrum features joint distribution adaptation. This method first uses Hilbert transform (HT) and fast Fourier transform (FFT) to construct the frequency domain envelope spectrum features of the raw signal. Then, in the cross-component fault diagnosis network, a new envelope spectrum joint distribution adaptation mechanism is introduced. By adjusting the weights of marginal distribution adaptation and conditional distribution adaptation, the domain alignment of the envelope spectrum between different components is achieved, and the fault diagnosis model's feature learning ability of the fault diagnosis model for faulty components in the target scenario is improved. Experiments show that the built cross-component transfer fault diagnosis network via envelope spectrum features joint distribution adaptive diagnosis accuracy of up to 93.5% on cross-component fault diagnosis task under the unlabeled conditions of target domain sample, proving the superior performance of the cross-component transfer fault diagnosis method proposed in this paper.